Biometrics, as a new emerging personal identification method, has received extensive attention by overcoming the disadvantage of traditional human identification techniques. Iris pattern is considered as one of the most promising biometrics modalities because of its uniqueness, stability, non-intrusiveness, and high reliability. Iris image preprocessing is important to iris recognition system, and significantly influences the performance of the system. This thesis investigates some key stages of iris preprocessing, with the purpose of improving system accuracy and reliability. Moreover, since most of the current iris recognition algorithms are evaluated on small databases, and collecting large iris image database is expensive in time and resource, another work of this thesis is to conduct research on iris image synthesis, with the purpose of providing iris image data for evaluating and comparing various algorithms. The main contributions of our work reported in this thesis are as follows: 1. A novel algorithm for iris image quality assessment has been proposed. We use three measures, which are spatial high frequency filters, vertical high frequency filters and average grey level of ROI, respectively, to discriminate defocused, motion blurred and occluded images. Our methods can make iris recognition system more robust to noise. 2. We have proposed two novel methods to correct nonlinear iris deformation. The first method is based on the assumption of iris structure. We use Gaussian model to describe the nonlinear deformation of iris texture. The second method is based on the appearance of iris texture, using local blob matching to evaluate iris deformation. The methods proposed in our work are effective, and iris recognition system with deformation correction is more robust in non-ideal environment. 3. We have proposed three measures for detecting counterfeit iris wearing printed color contact lens. The first method use iris edge sharpness as feature to detect counterfeit iris. Then we propose another method to learn a small finite vocabulary of micro-structures, which are called Iris-Textons, based on multi-channel Gabor filtering and machine learning. Then Iris-Texton histogram is used as feature vectors of iris textures. The third method uses features based on grey level co-occurrence matrix, and SVM as the final classifier. These methods perform well in detecting counterfeit iris, making the systems more robust in anti-spoofing. 4. We conduct resea...
修改评论